You may not have heard of Imgur, but the site serves 5 million images a day to an audience of 130 million visitors a month. And its popularity among viral meme makers is both its strength and its weakness.
“It’s almost like, before, one big firehose of images came through the site,” said CEO Alan Schaaf. “There was no way to slice it—you just got bombarded with everything.”
Schaaf is changing that. On Thursday morning, Imgur (pronounced “image-er”) is unveiling a new tagging system—and putting you to work solving its biggest problem.
Tags—simple, topical text labels—for online images are nothing new. The way Flickr put its community of photographers to work labeling each others’ work was a major reason why Yahoo bought it back in 2005.
Imgur users can now categorize all the images that make their way onto the site by tagging them. The most relevant tags for an image will be decided by consensus. And once those tags are in place, users will be able to track and follow a tag of choice—just like you do on Tumblr, Pinterest and Twitter.
(On some sites, tags are known as “hashtags” because the convention there is to use a hash mark—”#”—before a tag. Imgur isn’t using hashtags.)
Where Imgur does Tumblr and other sites that offer tag-based feeds one better is its combinations of tags. Where Tumblr might let you follow “cute” or “kitten,” Imgur lets you track those tags together, so you can see images labeled with both. You can also filter out tags you don’t want to see.
For example, Schaaf has a custom Game of Thrones gallery that tracks “got,” “Game Of Thrones,” and each character’s name. He knows the results he gets will be accurate, since Imgur users are voting on the tags. Meanwhile, another user might want to filter out these tags in order to avoid spoilers from the latest episode. Now, each user can choose to sample the gush of the Imgur firehose however he or she prefers
This human-powered way of organizing the Web, dubbed a “folksonomy” a decade ago, stands in contrast to the machine-driven image analysis companies like Google and Facebook are known for.
Imgur, long known for its stripped-down approach to image uploading and viewing, isn’t doing anything groundbreaking. But its choice to pursue people power over the increasingly advanced computational approaches available to it is interesting.
When asked to explain, Schaaf gave me the example of a black, red, and yellow layered cocktail. It was a popular image when Imgur tested tagging with a small group of users. Most testers, recognizing the festive drink’s connection to German soccer fans, correctly tagged it with “Germany” and “World Cup.”
“It’s about Germany and soccer as well as drinks, but a computer would only recognize the drink part,” he said. “It would never guess it’s related to the World Cup.”
Since Schaaf founded the site roughly five years ago out of his dorm room, the site has found a lot of popularity with users of Reddit, another site where users vote to find the most interesting articles and images online.
Imgur’s learning from Reddit’s flaws—for example, the way the site forces people to pick a single section, or subreddit, of the site when posting a link. Tim Hwang, director of special projects at Imgur, said it was important to find a tagging solution that would unite the community instead of fragment it. With multiple tags, images can fit into to several galleries instead of just a single designated one.
“We could have a ‘cat images’ category, but then people would identify with the ‘cat’ subgallery, or start forming rules around certain subgalleries that specified what kind of cat pictures were OK,” he said. “Tagging is something users can all do collectively.”
All Visual Web sites risk image overload. Without some way to discover specific images, like Pinterest’s guided search or Imgur’s new tags, users will only see the most popular stuff. That leads to a winner-take-all scenario where only images that make the homepage get seen. Schaaf’s goal is to help users expose the “dark matter” of Imgur that might get lost in the deluge.
“Whether they want to browse everything at once and filter out a couple of things, or drill down into a specific topic, people will make their own experience,” he said.